Identifying Latent Disease Factors Expressed in Subgroups of Patients with Depressive Disorders using Sparse Group Factor Analysis
Year: 2024
Type: dissertation
Abstract: This study applies a Sparse Group Factor Analysis (GFA) model with regularized horseshoe priors to a multimodal dataset of patients with major depressive disorder (MDD), including structural MRI, EEG recordings, behavioral tasks (Hariri Task and FERT), and question- naires. The primary goals were to identify latent factors distinguishing patients in remission from healthy controls, di!erentiate those who relapse from those who do not, and uncover subgroups within the broader depressed population. The Sparse GFA model successfully cap- tured both inter-modality and intra-modality variance, achieving group-wise, element-wise, and sample-wise sparsity. However, the dominance of EEG data posed challenges for captur- ing meaningful dependencies across other modalities. Comparative analysis with vanilla and non-regularized sparse GFA models demonstrated that regularized horseshoe priors enhance sparsity and interpretability. Notably, a latent factor dominated by beta band activity in cen- tral and parietal regions was di!erentially expressed between healthy controls and patients in remission (p = 0.028). In the study focusing on patients in remission, several latent dimensions were identified, including those reflecting grey matter variations, emotional visual processing, and subcortical-cognitive interactions. Additionally, two key axes emerged, representing the spectrum of general mental health and psychological well-being versus psychological distress. However, factors that could clearly di!erentiate between patients who relapse and those who do not remained elusive. These findings highlight the potential of multimodal data integration to improve diagnostic precision in depression and suggest further refinement of factor modeling strategies is needed.
Author: Enbiyaoglu, Mert
Language: English
Subfield: Pharmacology
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W6949637675
Cites:
Cited by:
Related to: